U.S. patent number 10,176,320 [Application Number 15/830,299] was granted by the patent office on 2019-01-08 for using machine learning in an industrial control network to improve cybersecurity operations.
This patent grant is currently assigned to Honeywell International Inc.. The grantee listed for this patent is Honeywell International Inc.. Invention is credited to Ganesh P. Gadhe, Kevin McMurdie.
![](/patent/grant/10176320/US10176320-20190108-D00000.png)
![](/patent/grant/10176320/US10176320-20190108-D00001.png)
![](/patent/grant/10176320/US10176320-20190108-D00002.png)
![](/patent/grant/10176320/US10176320-20190108-D00003.png)
United States Patent |
10,176,320 |
McMurdie , et al. |
January 8, 2019 |
Using machine learning in an industrial control network to improve
cybersecurity operations
Abstract
This disclosure provides a security system and method for using
machine learning to improve cybersecurity operations in an
industrial control networks and other systems. A method includes
collecting, by a security system, current process information for a
plurality of processes in a control system. The method includes
analyzing, by the security system, the current process information
according to one or more process models. The method includes
producing, by the security system and according to the analysis, a
risk report that identifies an abnormal process among the plurality
of processes.
Inventors: |
McMurdie; Kevin (Surprise,
AZ), Gadhe; Ganesh P. (Pune, IN) |
Applicant: |
Name |
City |
State |
Country |
Type |
Honeywell International Inc. |
Morris Plains |
NJ |
US |
|
|
Assignee: |
Honeywell International Inc.
(Morris Plains, NJ)
|
Family
ID: |
64815682 |
Appl.
No.: |
15/830,299 |
Filed: |
December 4, 2017 |
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G06F
21/577 (20130101); G06Q 10/0635 (20130101); G05B
13/0265 (20130101); G06F 21/552 (20130101); G05B
13/04 (20130101); G06F 21/56 (20130101); G05B
15/02 (20130101) |
Current International
Class: |
H04L
29/06 (20060101); G06F 21/55 (20130101); G06F
21/56 (20130101); G06F 21/57 (20130101); G06Q
10/06 (20120101); G05B 13/02 (20060101); G05B
13/04 (20060101); G05B 15/02 (20060101) |
Field of
Search: |
;726/22 |
References Cited
[Referenced By]
U.S. Patent Documents
Other References
Gerhards, "The Syslog Protocol", Network Working Group, Adiscon
GmbH, Mar. 2009, 38 pages. cited by applicant .
Knapp, "Security Intelligence and Analytics in Industrial Systems",
Honeywell Users Group Europe, Middle East and Africa, 2014, 29
pages. cited by applicant .
Koelemij, "Effective Defense in Depth Strategies for Industrial
Control Systems", Honeywell Users Group Europe, Middle East and
Africa, 2014, 21 pages. cited by applicant .
Koelemij, "Designing a Secure Industrial Control System", Honeywell
Users Group EMEA, Industrial IT Solutions, 2013, 11 pages. cited by
applicant .
Koelemij, "Making Cyber Security S.M.A.R.T.", Honeywell Industrial
Cyber Security EMEA, HUG Training Session, 2014, 63 pages. cited by
applicant .
Ricks, "System Center 2012 R2 Operations Manager Documentation",
Microsoft System Center 2012 R2, Microsoft Corp., Nov. 2013, 1389
pages. cited by applicant .
Spear, "Industrial Cyber Security 101", Honeywell Users Group
Europe, Middle East and Africa, 2015, 25 pages. cited by applicant
.
"Certification Testing Specification", OPC Foundation, Release
Candidate 1.02, May 2014, 20 pages. cited by applicant .
"Guide for Conducting Risk Assessments--Information Security", NIST
Special Publication 800-30, Revision 1, NIST, U.S. Dept of
Commerce, Sep. 2012, 95 pages. cited by applicant .
"Honeywell Industrial Cyber Security Overview and Managed
Industrial Cyber Security Services", Honeywell Process Solutions,
Jun. 2014, 19 pages. cited by applicant .
"Information Technology--Security Techniques--Information Security
Risk Management", Bio Standards Publication, BS ISO/IEC 27005,
2011, 80 pages. cited by applicant .
"Risk Management and Critical Infrastructure Protection: Assessing,
Integrating, and Managing Threats, Vulnerabilities and
Consequences", CRS Rpt. for Congress, RL32561, Feb. 2005, 28 pages.
cited by applicant .
"Security for Industrial Automation and Control Systems Part 2-3:
Patch Management in the IACS Environment", ISA-TR62443-2-3-2015,
Jul. 2015, 18 pages. cited by applicant .
"Design and Planning Guide for System Center 2012--Operations
Manager", System Center 2012, Microsoft Corp., Apr. 2012, 13 pages.
cited by applicant.
|
Primary Examiner: Paliwal; Yogesh
Claims
What is claimed is:
1. A method using machine learning to improve cybersecurity
operations in an industrial control networks comprising:
collecting, by a security system, current process information for a
plurality of processes in a control system, wherein collecting
comprises receiving source data; analyzing, by the security system,
the current process information according to one or more process
models, wherein the step of analyzing comprises: selecting a subset
of the source data, splitting the source data into first and second
data, filtering the first data to select normal data, tuning model
hyperparameters according to the filtered first data and a
one-class support vector model to produce a first model, producing,
by the security system and according to the analysis, a risk report
that identifies an abnormal process among the plurality of
processes, and wherein the security system comprises one or more
processing devices, one or more memories, and at least one network
interface.
2. The method of claim 1, further comprising: collecting, by the
security system, known process information for a plurality of known
processes executing in the control system; and augmenting, by the
security system, the one or more process models according to the
known process information.
3. The method of claim 2, wherein the security system collects the
known process information and augments the one or more process
models repeatedly, occasionally, or continuously.
4. The method of claim 1, wherein the current process information
includes a parent process name or a process name.
5. The method of claim 1, wherein the current process information
includes an execution path.
6. The method of claim 1, wherein the current process information
includes a session ID.
7. The method of claim 1, wherein the current process information
includes a number of instances of the process.
8. The method of claim 1, wherein the current process information
includes a startup time.
9. A security system to improve cybersecurity operations in an
industrial control networks, comprising: at least one network
interface; a hardware controller; and a memory in communication
with the controller, wherein the hardware controller is configured
to: collect current process information for a plurality of
processes in a control system wherein collecting comprises
receiving source data; analyze the current process information
according to one or more process models wherein the step of
analyzing comprises: selecting a subset of the source data,
splitting the source data into first and second data, filtering the
first data to select normal data, tuning model hyperparameters
according to the filtered first data and a one-class support vector
model to produce a first model; and produce, according to the
analysis, a risk report that identifies an abnormal process among
the plurality of processes.
10. The security system of claim 9, wherein the hardware controller
is further configured to: collect known process information for a
plurality of known processes executing in the control system; and
augment the one or more process models according to the known
process information.
11. The security system of claim 10, wherein the security system
collects the known process information and augments the one or more
process models repeatedly, occasionally, or continuously.
12. The security system of claim 9, wherein the current process
information includes a parent process name or a process name.
13. The security system of claim 9, wherein the current process
information includes an execution path.
14. The security system of claim 9, wherein the current process
information includes a session ID.
15. The security system of claim 9, wherein the current process
information includes a number of instances of the process.
16. The security system of claim 9, wherein the current process
information includes a startup time.
17. A non-transitory machine-readable medium encoded with
executable instructions that, when executed, cause one or more
controllers of a security system to: collect current process
information for a plurality of processes in a control system
wherein collecting comprises receiving source data; analyze the
current process information according to one or more process models
wherein the step of analyzing comprises: selecting a subset of the
source data, splitting the source data into first and second data
so that different or parallel processes can be executed, filtering
the first data to select normal data, tuning model hyperparameters
according to the filtered first data and a one-class support vector
model to produce a first model; and produce, according to the
analysis, a risk report that identifies an abnormal process among
the plurality of processes, wherein the security system comprises
one or more processing devices, one or more memories, and at least
one network interface.
18. The non-transitory machine-readable medium of claim 17, wherein
the non-transitory machine-readable medium is further encoded with
executable instructions that, when executed, cause the one or more
controllers of the security system to: collect known process
information for a plurality of known processes executing in the
control system; and augment the one or more process models
according to the known process information.
19. The non-transitory machine-readable medium of claim 18, wherein
the security system collects the known process information and
augments the one or more process models repeatedly, occasionally,
or continuously.
20. The non-transitory machine-readable medium of claim 17, wherein
the current process information includes a parent process name or a
process name, an execution path, a session ID, a number of
instances of the process, or a startup time.
Description
TECHNICAL FIELD
This disclosure relates generally to industrial systems. More
specifically, this disclosure relates to improving security for
industrial control networks and other networks.
BACKGROUND
New cyber security technologies that could be used in the
industrial control systems sector are slow to be adopted as the
technologies pose a risk that requires significant vetting from
each vendor. As a result, updates in security can lag changes in
threats. Improved systems are desirable.
SUMMARY
This disclosure provides a security system and method for using
machine learning to improve cybersecurity operations in an
industrial control networks and other systems. A method includes
collecting, by a security system, current process information for a
plurality of processes in a control system. The method includes
analyzing, by the security system, the current process information
according to one or more process models. The method includes
producing, by the security system and according to the analysis, a
risk report that identifies an abnormal process among the plurality
of processes.
Disclosed embodiments include a device or system comprising a
controller and a memory, configured to perform processes as
described herein. Disclosed embodiments also include a
non-transitory machine-readable medium encoded with executable
instructions that, when executed, cause one or more processors of a
system to perform processes as disclosed herein.
In various embodiments, the security system will also collect known
process information for a plurality of known processes executing in
the control system and augment the one or more process models
according to the known process information. In various embodiments,
the security system collects the known process information and
augments the one or more process models repeatedly, occasionally,
or continuously. In various embodiments, the current process
information includes a parent process name or a process name, an
execution path, a session ID, a number of instances of the process,
or a startup time.
Other technical features may be readily apparent to one skilled in
the art from the following figures, descriptions, and claims.
BRIEF DESCRIPTION OF THE DRAWINGS
For a more complete understanding of this disclosure, reference is
now made to the following description, taken in conjunction with
the accompanying drawings, in which:
FIG. 1 illustrates an example industrial process control and
automation system according to this disclosure; and
FIGS. 2 and 3 illustrate processes in accordance with disclosed
embodiments.
DETAILED DESCRIPTION
The figures, discussed below, and the various embodiments used to
describe the principles of the present invention in this patent
document are by way of illustration only and should not be
construed in any way to limit the scope of the invention. Those
skilled in the art will understand that the principles of the
invention may be implemented in any type of suitably arranged
device or system.
Emerging technologies pose a risk that requires significant vetting
from each vendor to ensure the technology does not introduce
control system instability. As a result, modern innovations in
cybersecurity mitigations lag adoption in the industry, creating a
security gap between process control systems and more traditional
computing environments. Disclosed embodiments include techniques
and solutions that address this gap and can specifically address
the unique challenges of the industrial control system
environment.
Disclosed embodiments can also use current security infrastructure
to address new and emerging threats. Due to the requirement for
stability in process control environments, changes to currently
supported and deployed antivirus solutions and operating system
patches must often be validated and qualified for a specific
environment. This requirement lengthens the gap between disclosure
of a vulnerability and remediation at the industrial control system
location, and this gap is a prime target for threat actors.
Disclosed embodiments can reduce or eliminate such delays.
FIG. 1 illustrates an example industrial process control and
automation system 100 according to this disclosure. As shown in
FIG. 1, the system 100 includes various components that facilitate
production or processing of at least one product or other material.
For instance, the system 100 is used here to facilitate control
over components in one or multiple plants 101a-101n. Each plant
101a-101n represents one or more processing facilities (or one or
more portions thereof), such as one or more manufacturing
facilities for producing at least one product or other material. In
general, each plant 101a-101n may implement one or more processes
and can individually or collectively be referred to as a process
system. A process system generally represents any system or portion
thereof configured to process one or more products or other
materials in some manner.
In FIG. 1, the system 100 is implemented using the Purdue model of
process control. In the Purdue model, "Level 0" may include one or
more sensors 102a and one or more actuators 102b. The sensors 102a
and actuators 102b represent components in a process system that
may perform any of a wide variety of functions. For example, the
sensors 102a could measure a wide variety of characteristics in the
process system, such as temperature, pressure, or flow rate. Also,
the actuators 102b could alter a wide variety of characteristics in
the process system. The sensors 102a and actuators 102b could
represent any other or additional components in any suitable
process system. Each of the sensors 102a includes any suitable
structure for measuring one or more characteristics in a process
system. Each of the actuators 102b includes any suitable structure
for operating on or affecting one or more conditions in a process
system.
At least one network 104 is coupled to the sensors 102a and
actuators 102b. The network 104 facilitates interaction with the
sensors 102a and actuators 102b. For example, the network 104 could
transport measurement data from the sensors 102a and provide
control signals to the actuators 102b. The network 104 could
represent any suitable network or combination of networks. As
particular examples, the network 104 could represent an Ethernet
network, an electrical signal network (such as a HART or FOUNDATION
FIELDBUS network), a pneumatic control signal network, or any other
or additional type(s) of network(s).
In the Purdue model, "Level 1" may include one or more controllers
106, which are coupled to the network 104. Among other things, each
controller 106 may use the measurements from one or more sensors
102a to control the operation of one or more actuators 102b. For
example, a controller 106 could receive measurement data from one
or more sensors 102a and use the measurement data to generate
control signals for one or more actuators 102b. Each controller 106
includes any suitable structure for interacting with one or more
sensors 102a and controlling one or more actuators 102b. Each
controller 106 could, for example, represent a
proportional-integral-derivative (PID) controller or a
multivariable controller, such as a Robust Multivariable Predictive
Control Technology (RMPCT) controller or other type of controller
implementing model predictive control (MPC) or other advanced
predictive control (APC). As a particular example, each controller
106 could represent a computing device running a real-time
operating system.
Two networks 108 are coupled to the controllers 106. The networks
108 facilitate interaction with the controllers 106, such as by
transporting data to and from the controllers 106. The networks 108
could represent any suitable networks or combination of networks.
As a particular example, the networks 108 could represent a
redundant pair of Ethernet networks, such as a FAULT TOLERANT
ETHERNET (FTE) network from HONEYWELL INTERNATIONAL INC.
At least one switch/firewall 110 couples the networks 108 to two
networks 112. The switch/firewall 110 may transport traffic from
one network to another. The switch/firewall 110 may also block
traffic on one network from reaching another network. The
switch/firewall 110 includes any suitable structure for providing
communication between networks, such as a HONEYWELL CONTROL
FIREWALL (CF9) device. The networks 112 could represent any
suitable networks, such as an FTE network.
In the Purdue model, "Level 2" may include one or more
machine-level controllers 114 coupled to the networks 112. The
machine-level controllers 114 perform various functions to support
the operation and control of the controllers 106, sensors 102a, and
actuators 102b, which could be associated with a particular piece
of industrial equipment (such as a boiler or other machine). For
example, the machine-level controllers 114 could log information
collected or generated by the controllers 106, such as measurement
data from the sensors 102a or control signals for the actuators
102b. The machine-level controllers 114 could also execute
applications that control the operation of the controllers 106,
thereby controlling the operation of the actuators 102b. In
addition, the machine-level controllers 114 could provide secure
access to the controllers 106. Each of the machine-level
controllers 114 includes any suitable structure for providing
access to, control of, or operations related to a machine or other
individual piece of equipment. Each of the machine-level
controllers 114 could, for example, represent a server computing
device running a MICROSOFT WINDOWS operating system. Although not
shown, different machine-level controllers 114 could be used to
control different pieces of equipment in a process system (where
each piece of equipment is associated with one or more controllers
106, sensors 102a, and actuators 102b).
One or more operator stations 116 are coupled to the networks 112.
The operator stations 116 represent computing or communication
devices providing user access to the machine-level controllers 114,
which could then provide user access to the controllers 106 (and
possibly the sensors 102a and actuators 102b). As particular
examples, the operator stations 116 could allow users to review the
operational history of the sensors 102a and actuators 102b using
information collected by the controllers 106 and/or the
machine-level controllers 114. The operator stations 116 could also
allow the users to adjust the operation of the sensors 102a,
actuators 102b, controllers 106, or machine-level controllers 114.
In addition, the operator stations 116 could receive and display
warnings, alerts, or other messages or displays generated by the
controllers 106 or the machine-level controllers 114. Each of the
operator stations 116 includes any suitable structure for
supporting user access and control of one or more components in the
system 100. Each of the operator stations 116 could, for example,
represent a computing device running a MICROSOFT WINDOWS operating
system.
At least one router/firewall 118 couples the networks 112 to two
networks 120. The router/firewall 118 includes any suitable
structure for providing communication between networks, such as a
secure router or combination router/firewall. The networks 120
could represent any suitable networks, such as an FTE network.
In the Purdue model, "Level 3" may include one or more unit-level
controllers 122 coupled to the networks 120. Each unit-level
controller 122 is typically associated with a unit in a process
system, which represents a collection of different machines
operating together to implement at least part of a process. The
unit-level controllers 122 perform various functions to support the
operation and control of components in the lower levels. For
example, the unit-level controllers 122 could log information
collected or generated by the components in the lower levels,
execute applications that control the components in the lower
levels, and provide secure access to the components in the lower
levels. Each of the unit-level controllers 122 includes any
suitable structure for providing access to, control of, or
operations related to one or more machines or other pieces of
equipment in a process unit. Each of the unit-level controllers 122
could, for example, represent a server computing device running a
MICROSOFT WINDOWS operating system. Although not shown, different
unit-level controllers 122 could be used to control different units
in a process system (where each unit is associated with one or more
machine-level controllers 114, controllers 106, sensors 102a, and
actuators 102b).
Access to the unit-level controllers 122 may be provided by one or
more operator stations 124. Each of the operator stations 124
includes any suitable structure for supporting user access and
control of one or more components in the system 100. Each of the
operator stations 124 could, for example, represent a computing
device running a MICROSOFT WINDOWS operating system.
At least one router/firewall 126 couples the networks 120 to two
networks 128. The router/firewall 126 includes any suitable
structure for providing communication between networks, such as a
secure router or combination router/firewall. The networks 128
could represent any suitable networks, such as an FTE network.
In the Purdue model, "Level 4" may include one or more plant-level
controllers 130 coupled to the networks 128. Each plant-level
controller 130 is typically associated with one of the plants
101a-101n, which may include one or more process units that
implement the same, similar, or different processes. The
plant-level controllers 130 perform various functions to support
the operation and control of components in the lower levels. As
particular examples, the plant-level controller 130 could execute
one or more manufacturing execution system (MES) applications,
scheduling applications, or other or additional plant or process
control applications. Each of the plant-level controllers 130
includes any suitable structure for providing access to, control
of, or operations related to one or more process units in a process
plant. Each of the plant-level controllers 130 could, for example,
represent a server computing device running a MICROSOFT WINDOWS
operating system.
Access to the plant-level controllers 130 may be provided by one or
more operator stations 132. Each of the operator stations 132
includes any suitable structure for supporting user access and
control of one or more components in the system 100. Each of the
operator stations 132 could, for example, represent a computing
device running a MICROSOFT WINDOWS operating system.
At least one router/firewall 134 couples the networks 128 to one or
more networks 136. The router/firewall 134 includes any suitable
structure for providing communication between networks, such as a
secure router or combination router/firewall. The network 136 could
represent any suitable network, such as an enterprise-wide Ethernet
or other network or all or a portion of a larger network (such as
the Internet).
In the Purdue model, "Level 5" may include one or more
enterprise-level controllers 138 coupled to the network 136. Each
enterprise-level controller 138 is typically able to perform
planning operations for multiple plants 101a-101n and to control
various aspects of the plants 101a-101n. The enterprise-level
controllers 138 can also perform various functions to support the
operation and control of components in the plants 101a-101n. As
particular examples, the enterprise-level controller 138 could
execute one or more order processing applications, enterprise
resource planning (ERP) applications, advanced planning and
scheduling (APS) applications, or any other or additional
enterprise control applications. Each of the enterprise-level
controllers 138 includes any suitable structure for providing
access to, control of, or operations related to the control of one
or more plants. Each of the enterprise-level controllers 138 could,
for example, represent a server computing device running a
MICROSOFT WINDOWS operating system. In this document, the term
"enterprise" refers to an organization having one or more plants or
other processing facilities to be managed. Note that if a single
plant 101a is to be managed, the functionality of the
enterprise-level controller 138 could be incorporated into the
plant-level controller 130.
Access to the enterprise-level controllers 138 may be provided by
one or more operator stations 140. Each of the operator stations
140 includes any suitable structure for supporting user access and
control of one or more components in the system 100. Each of the
operator stations 140 could, for example, represent a computing
device running a MICROSOFT WINDOWS operating system.
Various levels of the Purdue model can include other components,
such as one or more databases. The database(s) associated with each
level could store any suitable information associated with that
level or one or more other levels of the system 100. For example, a
historian 141 can be coupled to the network 136. The historian 141
could represent a component that stores various information about
the system 100. The historian 141 could, for instance, store
information used during production scheduling and optimization. The
historian 141 represents any suitable structure for storing and
facilitating retrieval of information. Although shown as a single
centralized component coupled to the network 136, the historian 141
could be located elsewhere in the system 100, or multiple
historians could be distributed in different locations in the
system 100.
In particular embodiments, the various controllers and operator
stations in FIG. 1 may represent computing devices. For example,
each of the controllers 106, 114, 122, 130, 138 could include one
or more processing devices 142 and one or more memories 144 for
storing instructions and data used, generated, or collected by the
processing device(s) 142. Each of the controllers 106, 114, 122,
130, 138 could also include at least one network interface 146,
such as one or more Ethernet interfaces or wireless transceivers.
Also, each of the operator stations 116, 124, 132, 140 could
include one or more processing devices 148 and one or more memories
150 for storing instructions and data used, generated, or collected
by the processing device(s) 148. Each of the operator stations 116,
124, 132, 140 could also include at least one network interface
152, such as one or more Ethernet interfaces or wireless
transceivers.
As discussed above, new products or processes to address security
issues can impact operations and as a result are often not
considered due to their large qualification costs leaving cyber
security operations to rely on network isolation as one of their
strongest security mechanisms.
Currently the majority of malware detection solutions employed in
the industrial control system (ICS) industry utilize known
signatures to detect malware. However, this is a reactive measure
against known non-polymorphic malware entities. An encrypted or
polymorphic variant that is not known to have a signature can
bypass existing antivirus (AV) countermeasures. For example, it can
be common for a new malware to have an undisclosed infection vector
or identifiable signature. Disclosed embodiments can aid in
identifying infection by behavior monitoring as disclosed herein.
Disclosed techniques provide a very effective response to these
types of malware campaigns and specifically address these threats
in a manner that is agnostic of the hardware implementation. Legacy
implementations are widespread in the ICS realm and solutions that
enable "bolt on" security are extremely valuable given the risks of
modifying legacy hardware/software within the ICS space.
Emerging and new variants also pose a unique threat in that they
can exploit the window during which AV signature files are released
by vendors and the actual update of a given ICS network's scheduled
update of those signature files. Disclosed embodiments provide
behavioral detection of a malware variant or threat actor, reducing
the risks between patch release and deployment.
The use of machine learning algorithms on valid data provides a
currently-nonexistent capability in the industrial controls
environment to identify system behaviors that could suggest a risk
to the system. Traditional cybersecurity software often introduces
overhead to a system that as a result disqualify them from use in
an ICS environment due to their resource intensive nature.
Disclosed embodiments, by contrast, can poll the needed data in a
periodic manner to minimize performance effects on a system. Since
disclosed embodiments do not calculate hashes of every file on a
system to compare to known antivirus signatures, as in conventional
systems, the resource consumption is insignificant and can be
offloaded to a separate node for further resource consumption
mitigation. This would be further relevant if employed on IIoT
devices with limited capabilities.
Industrial Internet of Things (IIoT) platforms are also subject to
emerging cybersecurity concerns and can also be subject to the same
limitations of being unable to support traditional cybersecurity
solutions due to operational impact or increased system overhead.
As IIoT solutions become increasingly targeted for cyber offensive
operations, new and innovative security solutions will be sought to
address security risks in a non-invasive manner.
Disclosed embodiments can be integrated or adopted in existing
products to augment their operations to offer an additional layer
of risk detection and mitigation. Disclosed systems and methods
provide risk analysis that traditional AV and other common
cybersecurity solutions cannot offer as they operate on real time
data and use known valid data originating from the site of
operation. As a result, disclosed embodiments have insight that
other security solutions do not leverage. This provides the ability
to identify new and emerging risks and potentially identify "zero
day" exploits targeting ICS and IIoT environments.
By analyzing known good data, machine-learning models employing
anomaly detection algorithms are trained to learn what is normal
and therefore detect abnormal behavior. As disclosed herein, if a
dataset exists and it is meaningful to examine any particular entry
for unusual data or behavior, then disclosed embodiments can be
tailored to address that question. As described above, machine
learning anomaly detection algorithms can be employed to a wide
range of tasks. While machine learning models can be trained using
live process data, in alternate embodiments, a valid baseline for
models can be crafted and deployed by the manufacturer given their
intimate knowledge of their own systems, and be used by processes
described herein. This can remove any need for running a system for
a period in order to establish a baseline.
Disclosed embodiments can analyze process information on computing
end nodes which can provide process data. Disclosed embodiments can
incorporate MICROSOFT WINDOWS process enumeration, LINUX process
enumeration and embedded device process enumeration. By training
models to learn what normal values are for monitored processes,
disclosed embodiments can identify abnormalities in near real time
(depending on poll times) and trigger an alert for
investigation.
Processes and results described herein can be accomplished (among
other ways) using a security system 154. Among other things, each
security system 154 can be implemented as one or more controller or
processor-based systems configured to perform processes as
disclosed herein.
In this example, each security system 154 includes one or more
processing devices 156; one or more memories 158 for storing
instructions and data used, generated, or collected by the
processing device(s) 156; and at least one network interface 160
(wired, wireless, or a combination of these). Each processing
device 156 could represent a microprocessor, microcontroller,
digital signal process, field programmable gate array, application
specific integrated circuit, or discrete logic. Each memory 158
could represent a volatile or non-volatile storage and retrieval
device, such as a random access memory or Flash memory. Each
network interface 160 could represent an Ethernet interface,
wireless transceiver, or other device facilitating external
communication. The functionality of the security system 154 could
be implemented using any suitable hardware or a combination of
hardware and software/firmware instructions. Database 155 can store
any data as described herein, including current process information
182, known process information 184, process model(s) 186, and a
risk report 188, as well as other process models, process data, and
other information as necessary, and can be implemented, for
example, in the one or more memories 158.
Processes disclosed herein can also or alternatively be accessed or
performed, in some cases, by an external system 170. In this
example, the external system 170 includes one or more processing
devices 176; one or more memories 178 for storing instructions and
data used, generated, or collected by the processing device(s) 176;
and at least one network interface 172. Each processing device 176
could represent a microprocessor, microcontroller, digital signal
process, field programmable gate array, application specific
integrated circuit, or discrete logic. Each memory 178 could
represent a volatile or non-volatile storage and retrieval device,
such as a random access memory or Flash memory. Each network
interface 172 could represent an Ethernet interface, wireless
transceiver, or other device facilitating external communication.
The functionality of the external system 170 could be implemented
using any suitable hardware or a combination of hardware and
software/firmware instructions. The external system 170 can be, for
example, a stand-alone data processing system, a mobile device, an
external server or enterprise system, or otherwise. The exemplary
structure of the external system 170 described above is not
intended to limit the structure or function of the devices that
could be used to implement the external system 170.
Although FIG. 1 illustrates one example of an industrial process
control and automation system 100, various changes may be made to
FIG. 1. For example, a control and automation system could include
any number of sensors, actuators, controllers, servers, operator
stations, networks, risk managers, and other components. Also, the
makeup and arrangement of the system 100 in FIG. 1 is for
illustration only. Components could be added, omitted, combined, or
placed in any other suitable configuration according to particular
needs. Further, particular functions have been described as being
performed by particular components of the system 100. This is for
illustration only. In general, control and automation systems are
highly configurable and can be configured in any suitable manner
according to particular needs. In addition, FIG. 1 illustrates an
example environment in which the functions of the security system
154 can be used. This functionality can be used in any other
suitable device or system.
FIG. 2 illustrates a process 200 in accordance with disclosed
embodiments for analyzing data for machine learning to improve
cybersecurity as disclosed herein. Such a process can be performed,
for example by a security system 154 or other system or elements as
disclosed herein (referred to generically as the "system"
below).
The system receives source data (202). The source data can be, for
example, a comma-separated-value (CSV) file or other file that
describes process information of processes that are executing or
have executed in a control system. The source data, depending on
the process, can act as current process information or known
process information.
The system selects a subset of the source data, such as specific
columns in a data set (204). The collected source data can be of a
known type or source to be analyzed according to the models
disclosed herein to identify changes or characteristics that are
abnormal to the host device. Any of the models referenced below can
be implemented as process model(s) 186.
The system can edit metadata for the source data (206). This
optional action can be performed to label or otherwise identify
specific data for further analysis, whether automated or
manual.
The system can split the source data into first and second data so
that different or parallel processes can be executed (208, 210).
This is not required to be an even split; for example, the system
could perform a 75%/25% stratified split on a Label column of the
source data. For purposes of description, assume that the first
data is the 75% split at 208 and the second data is the 25% split
at 210. Splitting the data can be used to provide a subset of data
used to train the models. Stratification can be used to ensure
samples are taken from both "good" and potentially "bad" datasets.
Of course, other data sampling or extraction techniques can be
used.
The first and second data can be further filtered to select, for
example, the rows with "normal" data that is not expected to show
any anomalies (212, 214). When training models, such filtering can
ensure that only known good data is fed into the model to enable it
to learn the characteristics of the good data.
The system can then tune model hyperparameters (216, 218). This
enables the model to be used with the most optimal hyperparameters.
This step creates and tests models and iteratively determines the
most optimal settings for the model.
For example, the system can tune model hyperparameters (216)
according to the filtered first data 212 and a one-class support
vector model 220 to produce a first model.
The system can train anomaly detection model (224) using the tuned
first model and the unfiltered first data. The system can then
score the first model (228).
In specific implementations, steps 224 and 228 can produce a
trained anomaly detection model. The tuned model hyper-parameters
can be fed into the trained anomaly detection model along with a
combination of good and bad data. This produces a score model that
can be used to evaluate the effectiveness of the trained anomaly
detection model.
The system can also process the second data. For example, the
system can tune model hyperparameters (218) according to the
filtered second data 214 and a PCA-based anomaly detection 222 to
produce a second model.
The system can train anomaly detection model (226) using the tuned
first model and the unfiltered first data. The system can then
score the first model (230).
The system can combine the first and second models into a combined
model, and evaluate the combined model (232). Alternately, where
there is only a single model as opposed to first and second models,
the system can evaluate the single model. This process can be used,
in particular, to train and validate models according to data known
to be "good" and data that reflects a potentially "bad"
condition.
By analyzing process information on a known valid dataset (the
filtered first and second data as discussed above forming known
process information), models can be trained to identify
abnormalities and notify an operator for further inspection. The
process information that can be analyzed and used to train these
models and to analyze processes can include a parent process name,
a process name (note that misspelled names suggest a mimicked
process), an execution path, a session ID, a number of instances of
the process, and a startup time, among other information.
By collecting such information to train machine learning models,
the models can learn the above attributes for a given ICS and/or
IIoT platform and offer insight when these values may suggest an
abnormality on the system. The values above are relatively
immutable in terms of their valid values and therefore are
potential indicators that a system may be at risk.
For malware detection, the trained models can determine a proper
execution path for an application or process. Models trained as
described herein can detect if a process's execution path is
modified or imitated. Execution paths can be used as a means to
detect if a process is being mimicked to appear to be a process
that would be expected on a given system and is an immutable
property of a given process as this does not change.
Disclosed embodiments can analyze parent processes. When a process
is spawned or created, the parent process for a given process is
always the same. This can be used to determine if a process on a
system has been tampered with or is attempting to mimic a known
valid process on a system.
Disclosed embodiments can analyze process image names. The process
image name is an immutable property of a process and can be used to
determine if a process has been created that is attempting to mimic
a known valid process.
Disclosed embodiments can analyze session IDs. Session ID is a
property used to determine the current session that generated and
owns the process. Typical operations in an ICS would allow a model
to identify expected session IDs for a given process as concurrent
sessions that would alter a process' session ID are not often
changed.
Disclosed embodiments can analyze the number of instances of a
process. The number of instances of a process can be viewed and
abnormalities in this number from the expected number of instances
could be used to raise awareness to a potential risk.
Disclosed embodiments can analyze an unusual startup time.
Collecting process startup times on an ICS and reviewing them can
provide a dataset to train machine learning models to identify
abnormalities.
Disclosed embodiments can analyze command lines. Each process entry
lists the command line entry that corresponds to the current
execution of the process. Select processes contain additional
parameters that can be supplied to machine learning models to
identify potential risks associated with abnormal parameter
entries.
FIG. 3 illustrates a process 300 in accordance with disclosed
embodiments that can be performed, for example, by a security
system 154 or other system, referred to generically as the "system"
below.
The system collects known process information for a plurality of
known processes executing in a control system (302).
The system builds or augments one or more process models according
to the known process information (304). Collecting the known
process information and building or augmenting the models can be
performed repeatedly, occasionally, or continuously to train the
models.
The system collects current process information for a plurality of
executing or recently executed processes in the control system
(306).
The system analyzes the current process information according to
the one or more process models (308).
According to the analysis, the systems produces a risk report that
identifies an abnormal process (310). The risk report can be in the
form of a warning screen or indicator, a text message or email, or
other electronic notification for use by a system operator or other
user.
In some embodiments, various functions described in this patent
document are implemented or supported by a computer program that is
formed from computer readable program code and that is embodied in
a computer readable medium. The phrase "computer readable program
code" includes any type of computer code, including source code,
object code, and executable code. The phrase "computer readable
medium" includes any type of medium capable of being accessed by a
computer, such as read only memory (ROM), random access memory
(RAM), a hard disk drive, a compact disc (CD), a digital video disc
(DVD), or any other type of memory. A "non-transitory" computer
readable medium excludes wired, wireless, optical, or other
communication links that transport transitory electrical or other
signals. A non-transitory computer readable medium includes media
where data can be permanently stored and media where data can be
stored and later overwritten, such as a rewritable optical disc or
an erasable memory device.
It may be advantageous to set forth definitions of certain words
and phrases used throughout this patent document. The terms
"application" and "program" refer to one or more computer programs,
software components, sets of instructions, procedures, functions,
objects, classes, instances, related data, or a portion thereof
adapted for implementation in a suitable computer code (including
source code, object code, or executable code). The term
"communicate," as well as derivatives thereof, encompasses both
direct and indirect communication. The terms "include" and
"comprise," as well as derivatives thereof, mean inclusion without
limitation. The term "or" is inclusive, meaning and/or. The phrase
"associated with," as well as derivatives thereof, may mean to
include, be included within, interconnect with, contain, be
contained within, connect to or with, couple to or with, be
communicable with, cooperate with, interleave, juxtapose, be
proximate to, be bound to or with, have, have a property of, have a
relationship to or with, or the like. The phrase "at least one of,"
when used with a list of items, means that different combinations
of one or more of the listed items may be used, and only one item
in the list may be needed. For example, "at least one of: A, B, and
C" includes any of the following combinations: A, B, C, A and B, A
and C, B and C, and A and B and C.
While this disclosure has described certain embodiments and
generally associated methods, alterations and permutations of these
embodiments and methods will be apparent to those skilled in the
art. Accordingly, the above description of example embodiments does
not define or constrain this disclosure. Other changes,
substitutions, and alterations are also possible without departing
from the spirit and scope of this disclosure, as defined by the
following claims.
* * * * *